Chipone Technology’s losses widened in the first half of the year, with the latest orders totaling over 3 billion yuan. It can provide over 40 TOPS computing power, and its NPU has already entered mass production and shipment.

① In the first half of this year, Xinyuan Technology Co., Ltd. achieved an operating income of 974 million yuan, an increase of 4.49% year-on-year; the net profit attributable to the parent company was -320 million yuan, which is a larger loss compared to the same period last year; ② Xinyuan Technology stated that driven by market growth, AI ASIC has become a necessity for high-efficiency AI computing, particularly experiencing rapid growth in the fields of ultra-large-scale data centers, real-time edge inference, and automotive systems.

According to the “Science and Technology Innovation Board Daily” on August 22 (Reporter Guo Hui), Xinyuan Technology announced its semi-annual report for 2025.

In the first half of this year, the company’s operating income was 974 million yuan, an increase of 4.49% year-on-year; the net profit attributable to the parent company was -320 million yuan, a larger loss compared to the same period last year; the net profit attributable to the parent company after deducting non-recurring gains and losses was -358 million yuan, compared to -304 million yuan in the same period last year; the net operating cash flow was -365 million yuan.

In the second quarter of this year, the company achieved an operating income of 584 million yuan, a decrease of 4.84% year-on-year, but an increase of 49.90% compared to the previous quarter; the net profit attributable to the parent company was -99.51 million yuan, a year-on-year decrease of 27.86%.

From a business perspective, in the first half of the year, Xinyuan Technology’s IP licensing revenue was 281 million yuan, an increase of 8.20% year-on-year; in the second quarter, IP licensing revenue was 187 million yuan, an increase of 99.63% quarter-on-quarter and an increase of 16.97% year-on-year. In the first half of the year, the company’s royalty revenue was 51 million yuan, roughly unchanged year-on-year.

In the first half of this year, Xinyuan Technology achieved revenue of 232 million yuan from chip design, a year-on-year decrease of 17.18%, with short-term fluctuations in revenue mainly affected by client project schedules. Among them, revenue from 28nm and below process nodes accounted for 89.39%, and revenue from 14nm and below process nodes accounted for 63.15%. In the first half of the year, revenue from chip design related to AI computing power accounted for approximately 52%.

In terms of volume business, revenue for the first half of the year was 408 million yuan, an increase of 20.47% year-on-year. In the second quarter, volume business revenue was 261 million yuan, an increase of 79.01% quarter-on-quarter and an increase of 11.65% year-on-year.

Regarding the order situation in the first half of the year, Xinyuan Technology signed new orders worth 1.656 billion yuan, an increase of 38.33% year-on-year, mainly for chip design and volume business orders. Among them, new orders for chip design amounted to 784 million yuan, an increase of 141.32% year-on-year; new orders for volume business were 665 million yuan, an increase of 39.60% year-on-year.

In the second quarter of this year, the company signed new orders amounting to 1.182 billion yuan, a nearly 150% quarter-on-quarter increase; the new order value for the chip design business exceeded 700 million yuan, with a quarter-on-quarter increase of over 700% and a year-on-year increase of more than 350%.

The launch of the DeepSeek large model has accelerated the rapid deployment of AI large language models across a wide range of edge devices. Chipone Technology stated that driven by market growth, AI ASICs have become a necessary option for high-efficiency AI computing, especially in rapidly growing fields such as ultra-large-scale data centers, real-time edge inference, and automotive systems.

The AI ASIC market is experiencing significant expansion. Marvell expects that the AI ASIC market size will reach approximately 55.4 billion USD by 2028; the customized chip market is projected to reach 94 billion USD by 2028. Broadcom anticipates that the customized AI chip market size could reach between 60 to 90 billion USD by 2027.

In terms of one-stop chip customization services, Chipone has design capabilities spanning advanced 5nm FinFET, 22nm FD-SOI to traditional 250nm CMOS processes, covering mainstream processes and special processes from major global foundries, and has successful tape-out experiences for chips at 14nm, 10nm, 7nm, 6nm, 5nm FinFET and 28nm, 22nm FD-SOI process nodes.

Chipone stated that among the total of 3.025 billion yuan in hand orders by the end of the second quarter of 2025, nearly 90% of the orders are for one-stop chip customization services, with an expected conversion rate of approximately 81% within one year. “As the company’s orders are gradually converted, the one-stop chip customization business will provide strong support for the company’s future revenue growth.”

Currently, Chipone’s neural network processor (NPU) IP has been used by 91 customers in over 140 artificial intelligence chips, and AI chips integrated with Chipone’s NPU IP have shipped nearly 200 million units globally.

It is understood that Chipone’s latest generation NPU architecture has been optimized for Transformer models, allowing it to efficiently run large language models such as Qwen and LLAMA, as well as support AIGC and multimodal models like Stable Diffusion and MiniCPM.

In the first half of this year, Chipone’s ultra-low power NPU has been able to provide over 40 TOPS of computing power for mobile large language model inference and has already been mass-produced in well-known companies’ smartphones and tablets. Chipone’s NPU is also deeply integrated with its numerous processor IPs, forming various AI acceleration subsystem solutions that include AI-ISP, AI-Display, AI-VPU, AI-GPU, and AI-DSP.

During the reporting period, VeriSilicon Holdings Co., Ltd. launched a scalable high-performance GPGPU-AI computing IP aimed at automotive and edge AI server applications, providing high compute density AI acceleration capabilities, multi-chip expansion support, and 3D stacked memory integration capabilities. The company is currently collaborating deeply with several leading AI computing clients to accelerate the large-scale implementation of these advanced technologies in practical applications.

VeriSilicon Holdings announced today (August 22) that six shareholders, including VeriSilicon Limited and Qionggang City Shixing Investment Partnership, intend to transfer 26,285,700 shares, accounting for 5.00% of the total share capital, through a price inquiry transfer method.